Theses and Dissertations

ORCID

0009-0006-4915-482X

Advisor

Thomasson, J. Alex

Committee Member

Zhang, Xin

Committee Member

Goodin, Christopher

Committee Member

Chen, Dong

Date of Degree

8-12-2026

Original embargo terms

Embargo 1 year

Document Type

Graduate Thesis - Open Access

Major

Biosystems Engineering

Degree Name

Master of Science (M.S.)

College

James Worth Bagley College of Engineering

Department

Department of Agricultural and Biological Engineering

Abstract

Recent research has emphasized reliable simulation environments for agricultural robotics. However, for simulation to support perception-driven autonomy, its sensor outputs must behave similarly to real-world sensing conditions. This thesis improves AG-MAVS, a high-fidelity agricultural simulation framework built on the Mississippi State University Autonomous Vehicle Simulator, by increasing stereo depth realism through learned adaptation using simulated agricultural data and real depth camera observations. The adapted depth outputs were evaluated in matched real and simulated agricultural scenarios using RTAB-Map and ORB-SLAM3. After real-field calibration, the model reduced End-Point Error from 80.37 to 7.56 pixels, an improvement of approximately 91%. Improved depth estimation also increased valid depth coverage from approximately 23%–38% to 77%–88%. These results indicate that the adapted AG-MAVS depth outputs more closely represent real agricultural sensing conditions and improve feature observability, loopclosure stability, and SLAM performance when degradation is caused by sparse or inconsistent depth. However, it cannot fully resolve synchronization errors, vegetation realism limitations, or other simulator-related discrepancies.

Sponsorship (Optional)

Mississippi Agricultural and Forestry Experiment Station (MAFES)

Download

Available for download on Thursday, September 09, 2027

Share

COinS